logbin,
logbin.allref takes the formula and data for a
log-link binomial GLM and produces a list of all
parameterisations needed for the associated CEM algorithm.
logbin.allref(object, data = environment(object), mono, start = NULL)model.frame. If another sort of object,
model.frame is called first.
start was specified, the first component for each
term will correspond to the parameterisation specified
by start.terms component of object.data
argument, or the result of calling model.frame with data.terms are restricted to be
monotonically non-decreasing.start, corresponding
to the first parameterisation in allref. NULL if start
was not supplied.logbin, the
parameter space is partitioned into a collection of
restricted parameter spaces (see Marschner, 2014).
logbin.allref finds the list of possible
parameterisations of each term in the model.If a term x has a TRUE value for
is.factor(x), is.character(x)
or is.logical(x), it is considered to be a
categorical covariate. This has a
parameterisation for each level of the factor.
Otherwise the covariate is considered to be continuous, in which case it has two possible parameterisations, relating to the minimum and maximum observed values.
If a covariate is restricted to be monotonic via the
mono argument, it has only one parameterisation.
logbin considers all possible combinations of
the parameterisations of each covariate, and for each calls
logbin.design to create the appropriate
non-negative design matrix to be used in the EM algorithm.
Marschner, I. C. and A. C. Gillett (2012). Relative risk regression: reliable and flexible methods for log-binomial models. Biostatistics 13(1): 179--192.
logbin